Skip to content
CodeVetter docs
Esc
↑↓navigate↵open⌘Jpreview
On this page

Test, qualification, and code-health surfaces for the native app and shared Rust core.

Testing

Tests are grouped by the boundary they prove. Passing one boundary never substitutes for another.

Native app

pnpm test:native:background

The background lane runs Swift package behavior tests, isolated native performance contracts, and a Debug Xcode build without app activation.

pnpm test:native:ui

The UI lane runs only XCUITest interaction targets. It requires a dedicated graphical desktop and explicit approval because it can move focus.

pnpm test:native:full

The full lane keeps background checks before UI automation. Pull requests use a hosted macOS runner so local work is not disturbed.

Rust core, CLI, and MCP

pnpm core:test
pnpm core:clippy
pnpm core:fmt
cargo test --manifest-path crates/codevetter-core/Cargo.toml --bin codevetter-mcp
cargo test --manifest-path crates/codevetter-core/Cargo.toml --test mcp_stdio
cargo test --manifest-path crates/codevetter-core/Cargo.toml --features browser-agent --bin codevetter

The CLI and MCP tests prove parsing, receipt semantics, repository scope, JSON stdio behavior, cancellation, pagination, and the read-only agent boundary.

Package and release contracts

pnpm test:native-runner
pnpm test:native-package
pnpm test:native-package-finalize
pnpm test:native-appcast
pnpm test:native-notarization
pnpm test:native-data-continuity
pnpm test:native-installed-upgrade
pnpm test:native-release

These are deterministic contract tests. The protected production workflow is still required for real Developer ID signing, Apple notarization, Sparkle signing, Gatekeeper, installed upgrade, data continuity, and rollback.

Repository automation

pnpm test:automation
pnpm test:corpus-contracts
pnpm test:retrieval
pnpm test:core-tools
pnpm capabilities:check

Code health and docs

pnpm lint
pnpm knip:strict
pnpm quality:complexity
pnpm quality:cycles
pnpm quality:duplication
pnpm quality:dependencies
node scripts/check-docs.mjs

The change-size and complexity gates compare against a base revision. CI fetches that exact base before running them.

Evidence interpretation

Large Repo Unpack data imports use a bounded corpus batch gate alongside the unchanged source-code change-size limits. A dated receipt must bind each addition by hash and source pin, preserve the baseline corpus and projections, and retain passing qualification evidence. The batch is capped at 100 additions and 128 MiB; generated projections must reproduce exactly. This is artifact admission, not independent semantic or runtime verification. Its rejection contracts run in test:automation alongside citation and skill-recorder tests.

The dated testing-speed study uses the Fleet skill trial to prioritize startup, selection, reuse, and evidence collection experiments. Its recommendations are unimplemented and require paired validation; they do not change these qualification requirements.

  • Node-test qualification uses the built-in runner’s structured completion events, retaining TAP output for metrics and failure frames. Named checks require one completed, unambiguous body; skips, TODOs, file-only wrappers, printed names, stale/incomplete reports and cancellation cannot establish execution. Performance execution receipts retain the bounded selection counts. These are local runtime observations, not authentication of externally supplied evidence.
  • A unit test proves its contract, not a production release.
  • A local package may be ad-hoc signed; shipping requires Developer ID and notarization receipts.
  • A successful build does not prove launch, interaction, upgrade, or rollback.
  • MCP remains read-only even when the native app and CLI can execute.
  • Missing provider or runtime evidence is unavailable, never zero or passing.

Was this page helpful?